Hadron v0.29.0

Added

• Cross-Platform Process Sandboxing (hadron-gatekeeper, hadron-gluon, hadron-chamber):
  • Implemented zero-overhead native sandboxing across Linux/WSL (Bubblewrap / bwrap), macOS (Apple Seatbelt / sandbox-exec dynamic SBPL Scheme profiles), and Windows (PowerShell containment harness & Win32 Job Objects with KILL_ON_JOB_CLOSE).
  • Added seat-level and team configuration (team.json / Team::sandbox_mode) supporting Strict, WorktreeOnly, and Off isolation policies with automatic worktree confinement and host credential environment scrubbing (AWS_*, GITHUB_*, ANTHROPIC_*, OPENAI_*).
  • Added Environment Settings card controls including "Enable process sandbox" toggle and confinement mode selector (Off / Worktree / Strict), plus active sandbox driver diagnostics.
  • Implemented interactive [SANDBOX: ON / STRICT / OFF] status badge in the chat input bar positioned between [Mode] and [@Quark] with click and F7 shortcut cycling.
• Frontier LLM Ergonomics & Efficiency Suite (hadron-gluon, hadron-forge-mcp):
  • Structured prompt generation into an immutable cache-stable prefix to maximize KV prompt cache hit rates (80%+ cost and latency reduction across Claude, Gemini, and OpenAI models).
  • Integrated automated transcript test runner output folding into SlidingContextPruner::prune_projection, persisting verbose logs to .hadron/scratch/ and keeping chat context concise.
  • Enabled multi-modal base64 screenshot piping in ACP turns, automatically loading referenced .hadron/screenshots/*.png files into ContentBlock::Image payloads.
  • Implemented batched parallel tool dispatch (dispatch_batched_tool_calls), evaluating concurrent tool calls in parallel over asynchronous futures in ACP and local model loops.
  • Wired speculative multi-model tournament candidate evaluation (speculative_tournament_winner) in hadron-gluon::bakeoff, prioritizing green merge gate verification and minimal diff churn.